Automatic Hip Fracture Identification and Functional Subclassification with Deep Learning.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Hip fractures are a significant cause of morbidity and mortality.
• Accurate and timely diagnosis is crucial for optimal patient outcomes.
• Current diagnostic methods can be subject to errors and delays.

Purpose of the Study:

• To assess the feasibility of using deep learning for automatic hip fracture identification and classification.
• To determine if deep learning can reduce diagnostic errors and expedite surgical intervention.
• To compare the performance of a deep learning model against human observers.

Main Methods:

• A deep learning object detection model was trained on 1118 hip and pelvic radiograph studies (3026 hips).
• The model utilized a Densely Connected Convolutional Neural Network (DenseNet) for classification.
• Performance was evaluated against fellowship-trained radiologists, orthopedists, and senior residents.

Main Results:

• The deep learning model achieved high accuracy in fracture detection (93.7%) and multiclass classification (90.8%).
• Model performance was comparable to expert human observers.
• AI assistance improved human diagnostic accuracy, particularly for residents.

Conclusions:

• Deep learning models demonstrate expert-level performance in identifying and classifying hip fractures.
• AI tools can serve as valuable aids to improve human diagnostic performance.
• This technology holds potential for enhancing patient care by reducing diagnostic errors and improving treatment timeliness.